AMD Radeon RDNA2 Navi (RX 6500, 6600, 6700, 6800, 6900 XT)

It could actually be a hard cap by default. Unlike Zen3, which boosts beyond advertised boost clock as long as you have a decent silicon and keep dropping the temp.

 

I guess you will have to wait and see then come November 18th.

 

I don't OC but looks like no restrictions for those who do:

 

With Navi 21 it looks like you're leaving a lot of performance on the table if you don't overclock, though.

 

Like in good ol' days

 

As long as it can do so while staying quiet. Whatever my next GPU is it's got to be near silent.

 

I definitely agree but that's an easier decision to make when there is plenty of stock to choose from and fair pricing. I feel like I'll just be buying reference and hoping it's a good cooler. At least it's not a blower. Also, aftermarket cards can be so overly inflated in price where I live. With the 3080 for example prices vary by $500AUD for custom boards, with a typical one costing at least $150-200AUD more than a unobtainable FE model...

The other issue is that I'd normally go for for something around 250w like a 6800 partly because of the lower power use, but the price difference needs to be worth it also. The $80 difference between it and the 6800XT is so small, I feel like it should have been $550.

At least the XT has a 2.5 slot cooler so in theory it will hopefully remain reasonably quiet with the higher clock speeds. I image a stock cooler 6800 overclocked would be pretty loud. I'll have to see what local prices are, but I'm thinking a 6800XT will be the better deal with USD MSRP price, and with more CU's and ROPS, it should do well even if power limited/underclocked a little for keeping the temps and noise in check.

 

That depends on wheter or not it's a 20% OC (in contrast to a 5-10% OC).

 

Have you considered a waterblock ?

 

You mean we'll need to see where the average clocks actually stand with these cards whole gaming, as the "game clock" values might not represent that average anymore?

 

If your system setup allows for effective cooling and power, your going to average above those game clocks. On dGPU's you get the minimum clocks as advertised, but much higher if temperatures and power supply etc allow.
In NV's case, your almost always at much higher then the advertised clocks/performance numbers. A 2080Ti is advertised as a 13TF gpu but actually spends its time at 16TF for about anything.

 

Teraflops isn't the right term to use when talking about game performance.

 

Game clocks in AMD's parlance is that it's supposed to be a conservative baseline for clock speeds in gaming work loads (specifically gaming, base clock is still the actual base line). But AMD's been more aggressively exploring clocking/gating for both technical reasons and marketing reasons as of late so who knows if they change how it functions for this generation.

AMD's boost clock is an actual limit which is different than how Nvidia uses it (not sure if Nvidia has a universal limit for a model, but they don't advertise one in spec. Nvidia's own "boost clock" is an even more conservatives rating than AMD's "game clock").

As aside since TFLOPs was brought up something to keep in mind is for the advertised "stock" TFLOPs spec is while both AMD and Nvidia calculates it based on the official "boost clock" their usage of the term is completely different. This is in addition to the apples/oranges issue of comparing TFLOPs to effective performance with differing uarchs.

On a related note I'm still of the mindset that TFLOPs is really just a easier to work with number in place of clock speed x FPUs (as the numbers involved are huge by comparison), and that's really all it should be.

 

This generation it seems the number of “cores” (SM/CU) will be a much better proxy for gaming performance than raw TFLOPs.

 

I'm not sure I follow? Number of cores is just one half of the equation. You need to know how fast those cores are running to get an idea of the potential performance. 50 cores at 2 Ghz will be much faster than 70 cores at 500Mhz for example.

As arandonguy said, TFLOPS is just a convenient measure for taking account of both of those factors.

You also need to look at front end throughput, and memory bandwidth. And that's just within the same architecture.

 

Based on what we know so far TFLOPs is a terrible predictor of performance with the new cards so whats the point of using it?

I’m not saying number of SMs or CUs is a perfect proxy for actual performance. I’m saying it’s a much better one than TFLOPs.

 

But TFLOPS is a factor of the number of cores? Cores * Clock Speed = TFLOPS. So how can no of cores be a better indicator than TFLOPS? I don't disagree that looking at TFLOPS alone isn't a good idea. But looking at it as part of a larger picture (bandwidth, front end performance, the demands of the particular game/scenario) can be useful. But it's certainly not the be all and end all.

 

A Finnish etailer Multitronic has become the first to list upcoming custom Radeon RX 6800 series.

https://videocardz.com/newz/sapphire-radeon-rx-6800-xt-nitro-series-listed-for-preorder-in-finland

 

In Cerny's words, A rising tide ...